doi:10.

1093/brain/awr259

Brain 2011: 134; 36873696

| 3687

BRAIN
A JOURNAL OF NEUROLOGY

The balance between cognitive reserve and

brain imaging biomarkers of cerebrovascular
and Alzheimers diseases
Alison D. Murray,1 Roger T. Staff,1,2 Christopher J. McNeil,1 Sima Salarirad,1 Trevor S. Ahearn,1
Nazahah Mustafa1 and Lawrence J. Whalley1

Correspondence to: Alison D. Murray,

Aberdeen Biomedical Imaging Centre,
Lilian Sutton Building,
Foresterhill,
Aberdeen AB25 2ZD, UK
E-mail: a.d.murray@abdn.ac.uk

The cognitive reserve hypothesis explains the disparity between clinical and pathological phenotypes and why, in two individuals with the same extent of neuropathology, one may be demented while the other remains cognitively intact. We examined
the balance between brain magnetic resonance imaging measures of the two most common pathologies associated with brain
ageing, cerebrovascular disease and Alzheimers disease, and parameters of cerebral reserve in well-characterized participants
born in 1936, for whom childhood intelligence is known. Brain magnetic resonance imaging was carried out at 1.5T using fluid
attenuation inversion recovery and T1-weighted volumetric sequences in 249 participants. Cerebrovascular disease was quantified by measuring brain white matter hyperintensities on fluid attenuation inversion recovery images using Scheltens scale and
Alzheimers disease was measured from volumetric data using FreeSurfer to extract whole brain volume and hippocampal
volumes in turn. The effect of these measures of brain burden on life-long cognitive ageing from the age of 11 to 68 years
was compared with the effect of educational attainment and occupational grade using structural equation modelling. Complete
brain burden and reserve data were available in 224 participants. We found that educational attainment, but not occupation, has
a measurable and positive effect, with a standardized regression weight of + 0.23, on late life cognitive ability in people without
cognitive impairment aged 68 years, allowing for the influence of childhood intelligence and the two most common subclinical
brain pathological burdens in the ageing brain. In addition, we demonstrate that the magnitude of the contribution of education
is greater than the negative impact of either neuropathological burden alone, with standardized regression weights of 0.14 for
white matter hyperintensities and  0.20 for hippocampal atrophy. This study illustrates how education counteracts the deleterious effects of cerebrovascular disease and Alzheimers disease and highlights the importance of quantifying cognitive reserve
in dementia research.

Keywords: reserve; white matter hyperintensity; cohort study; cognitive decline; hippocampus

Received March 21, 2011. Revised July 5, 2011. Accepted July 17, 2011. Advance Access publication November 18, 2011
The Author (2011). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved.
For Permissions, please email: journals.permissions@oup.com

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

1 Aberdeen Biomedical Imaging Centre, College of Life Sciences and Medicine, University of Aberdeen, Kings College, Aberdeen AB24 3FX, UK
2 NHS Grampian, Foresterhill, Aberdeen AB25 2ZD, UK

3688

| Brain 2011: 134; 36873696

Introduction

As well as being an important confounder of cognitive outcome

variables, understanding the mechanisms that permit some individuals to live independent lives in the presence of brain pathology
possesses considerable societal value (EClipSE Collaborative
Members, 2010).
The two most common neuropathologies of ageing are
Alzheimers disease and cerebrovascular disease (Matthews
et al., 2009), both of which can cause cognitive impairment and
dementia. Recognized imaging features of these are whole brain
and hippocampal atrophy and white matter hyperintensities, respectively, both of which can be measured on brain magnetic
resonance imaging (Manolio et al., 1994; Dubois et al., 2007).
We have previously studied reserve in a sample of healthy
78-year olds and showed that cognitive ability in late life, accounting for childhood cognitive ability and brain burden (whole brain
atrophy and white matter hyperintensities), is not related to total
intracranial volume (as an estimate of original brain size), but is
related to duration of education and occupational complexity,
indicating that, in this sample, reserve is likely to be active
rather than passive (Staff et al., 2004). Here, we investigate the
balance between measures of cognitive reserve and quantifiable
brain imaging biomarkers of neuropathology in cognitive ageing.
We studied a well-characterized, non-demented sample using an
exploratory structural equation modelling approach in order to
investigate the interrelations between these positive and negative
influences.

Subject and methods

Participants
The Scottish Mental Survey of 1947 tested the intelligence of all
children (n = 75 211) attending school, aged 11 years and is described elsewhere (Deary et al., 2004). Aged 64 years, 506 local
survivors, known as the Aberdeen Birth Cohort of 1936, were
recruited into a longitudinal study of brain ageing and health,
including biannual cognitive testing. From 20035, aged 679
years, 249 of these participants gave informed consent to brain
MRI and met inclusion criteria: living independently in the community, absence of dementia or known neurological disease and
usual contraindications to MRI, such as presence of a cardiac
pacemaker or other implanted metallic device. This study was
approved by the Local Research Ethics Committee.

Cognitive testing
Cognitive ability aged 11 years was measured using the Moray
House Test No. 12, which is a valid measure of intelligence and
has been shown to correlate (0.8) with the StanfordBinet
Intelligence Quotient test. A psychologist administered the following neuropsychological tests in later life: Ravens Progressive
Matrices (Raven et al., 1977), a non-verbal reasoning measure of
fluid intelligence; the Auditory Verbal Learning Test (Rey, 1964) a
measure of immediate and delayed verbal memory; the Uses of
Common Objects test, a measure of executive function or purposive action (Guilford et al., 1978) and the Digit Symbol test

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

The cognitive reserve hypothesis explains the disparity between

clinical and pathological phenotypes and why, in two individuals
with the same extent of neuropathology, one may be demented
while the other remains cognitively intact and possesses sufficient
reserve to balance their neuropathological change brought about
by ageing and disease. In this study we shall refer to these
changes as burden. The reserve hypothesis can be applied to
many brain diseases, mainly in the fields of cognitive ageing, cognitive impairment and dementia, and also to head injury, schizophrenia, depression and multiple sclerosis (Sumowski et al., 2010).
How reserve is acquired and implemented is unclear. Stern (2002)
has hypothesized that such reserve may be passive or active.
The former implies that reserve is provided by having a bigger
brain, which protects against the destructive effects of neuropathology, the latter that redundant networks are able to take over the
function of damaged networks, maintaining cognitive ability in the
face of deleterious pathology. Emerging evidence supports both
active cognitive reserve, which may allow the brain to cope with
pathology by use of pre-existing cognitive processes (neural reserve) or recruitment of compensatory networks (neural compensation) (Stern, 2009), and passive or brain reserve provided by a
bigger brain or larger volumes of specific regions. The passive
model of reserve does not account for individual differences in
cognitive processing. Of course, the different models of reserve
are not mutually exclusive and commonly used parameters of reserve, such as educational duration and attainment are likely to
underlie both brain and cognitive (neural) reserve.
Many factors influence late life cognitive ability and these can
be broadly regarded as a balance between positive and negative
factors. The principal positive influence is childhood intelligence
and it is recognized that this predicts 4050% of the variance in
fluid intelligence in the eighth decade (Deary et al., 2003). Higher
childhood intelligence will contribute to reserve through greater
uptake of educational opportunities, leading to more complex occupations and diversity of leisure activities (Gold et al., 1995).
Childhood intelligence is likely to be the result of a combination
of genetic and early life environmental factors. An adverse intrauterine environment resulting in low birth weight increases the risk
of vascular disease in later life, although the precise mechanism of
this increased risk is unknown. However, not all the variance in
cognitive ability in late life can be explained by measurable positive and negative factors and cognitive reserve can be used to
describe that proportion of variance unexplained by the opposing
effects of measurable pre-morbid ability and neuropathology or
brain burden. Such cognitive reserve protects against the clinical
manifestation of dementia-related neuropathology: those with
more reserve present with more advanced disease (Mortimer
et al., 2005). However, once the clinical threshold is reached,
those with greater reserve progress fasterin other words the
point of inflection is later in those with more reserve (Hall et al.,
2007). Intuitively, this means that those with greater reserve are
less amenable to early detection using cognitive measures
(Querbes et al., 2009). It is critical in cognitive declinedementia
research and in clinical trials to understand and quantify reserve.

A. D. Murray et al.

Cognitive reserve and neuroimaging disease burden

(Wechsler, 1997), a measure of processing speed, attention and
visual short-term memory.

Brain magnetic resonance imaging

acquisition and analysis

Cognitive reserve parameters

Estimates of how the brain had been used during adult life were
maximal educational attainment, based on qualification achieved
and maximal occupational grade. Values for educational attainment ranged from no qualification (Score 1) to completion of a
higher or professional degree such as a Doctor of Philosophy or
Doctor of Medicine (Score 9). In detail, those with no qualifications were coded 1; those with a lower leaving certificate coded 2;
those with a higher leaving certificate coded 3; Scottish vocational
certificate holders were coded 4; Ordinary levels/Standard grades
were coded 5; those achieving Higher/Advanced Levels were coded
6, participants with undergraduate or postgraduate (Masters) qualifications were coded 7 and 8, respectively, and those with a
higher or professional degree were coded 9, Occupational grade
was recorded as coded by the UKs Office of Population Statistics
classification (Great Britain Office of Population Censuses and
Surveys, 1990). This classifies occupational status as follows; 1:
managerial, 2: professional, 3: lesser professional, 4: secretarial,

| 3689

5: skilled manual, 6: semi-skilled ii, 7: semi-skilled i, 8: unskilled

ii and 9: unskilled i. These categories are the best occupational
level attained by an individual. This occupational classification gives
higher status occupations, often with more complex cognitive demands, a low score and places unskilled manual workers in the
highest classification. For the purposes of this analysis, the scale of
occupational grades was inverted.

Statistical analysis and modelling

Bivariate (Pearsons) correlations were used to examine relationships between variables. Structural equation models were constructed to test the hypothesis that childhood intelligence,
education and occupation would have a positive influence on
late life intelligence, while brain imaging measures of subclinical
cerebrovascular and Alzheimers disease would have a negative influence. Neuropsychological tests in late life are all highly and
positively inter-correlated. It is possible that factors of reserve
and burden may influence discrete cognitive domains differently.
One statistical approach would be to obtain an estimate of g as a
latent variable within the structural equation model by including all
of the raw cognitive scores. This would allow exploration of direct
effects of measures of burden and reserve on separate cognitive
domains not explained by g. Although possible, with the number
of participants and data points presented here this would have
resulted in a significantly less parsimonious model, open to the
criticism of over-fitting. For this reason, we chose to reduce our
cognitive data using principal components analysis prior to modelling. We constructed two structural equation models using AMOS
18. In the first we hypothesized that childhood intelligence would
have a direct effect on late life ability g, education and occupation.
Education and occupation were also hypothesized to have a direct
effect on g. Sex was hypothesized to have a direct effect on g,
whole brain atrophy and total intracranial volume. White matter
hyperintensities, total intracranial volume and whole brain atrophy
were hypothesized to have a direct influence on g. Hoyle and
Panter (1995) guidelines were used to assess the goodness of fit
for resulting models according to the following criteria: chisquared to degrees of freedom ratio close to 1, incremental fit
index 40.95 and the comparative fit index 40.95. In addition,
the root mean square error of approximation is reported, where a
value 50.05 indicates a good fit, and 50.08 an acceptable fit.

Results
Participants
Table 1 shows a comparison between participants in this brain
imaging study, compared with Aberdeen 1936 Birth Cohort members who did not participate. Participants had higher childhood
intelligence, late life Ravens Progressive Matrices and Digit
Symbol scores. In addition, females had significantly higher
(P 5 0.001) scores on the Auditory Verbal Learning Test and
males had significantly greater head and brain volumes (total
intracranial volume, P 5 0.001; brain volume, P 5 0.001; hippocampal volumes, P 5 0.001). There were no sex significant

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

Participants were imaged on a 1.5T General Electric NVi system

(General Electric) using axial T2-weighted, fluid attenuation inversion recovery and 3D T1 spoiled gradient recalled (SPGR) sequences with the following parameters: T2-weighted repetition
time/echo time 4900/81.4 ms, slice thickness 5 mm, space
1.2 mm; fluid attenuation inversion recovery repetition time/echo
time 9002/1.33 ms, inversion time 2200 ms, slice thickness 5 mm,
space 1.2 mm; and 3D T1 repetition time/echo time 20/6 ms, flip
angle 35 , number of slices 124, effective slice thickness 1.6 mm,
matrix 256  92, in-plane resolution 1  1 mm. White matter
hyperintensities were assessed using Scheltens scale (Scheltens
et al., 1993) in the following regions: frontal, parietal, temporal
and occipital white matter (the total of which gave the white
matter hyperintensity score); periventricular around the frontal
horns, bodies and occipital horns of the lateral ventricles (the
total of which gave the periventricular hyperintensity score);
deep grey matter of the caudate, putamen, thalamus and hippocampus (the total of which gave the grey matter hyperintensity
score); and infratentorial regions of the midbrain, pons, medulla
and cerebellum (the total of which gave the infratentorial hyperintensity score) by an experienced blinded observer. The sum of
scores in these four regions gave a total Scheltens score and this
value was entered into the analysis. The T1-weighted images were
exported and processed offline on a desktop computer running
Scientific Linux (version 5.5). Volumetric segmentation was performed with the FreeSurfer image analysis suite, which is documented and freely available for download online (http://surfer.
nmr.mgh.harvard.edu/). The left and right hippocampal volumes
and total brain volume were extracted from the FreeSurfer output
files automatically for each participant. The reciprocal of these
volumes was used in analyses as an estimate of atrophy.

Brain 2011: 134; 36873696

| Brain 2011: 134; 36873696

3690

A. D. Murray et al.

differences in other cognitive variables, in educational attainment

or occupational grade.

Magnetic resonance imaging variables

Most participants had some white matter hyperintensities, with total
Scheltens scores ranging from 047 (mean 16.5). Most white matter
hyperintensities were in the frontal lobes. Visual rating scales are
subject to inter-observer variation. A subset of 20 image data sets
was scored by a second experienced observer. The correlation
between observers was r = 0.77 indicating substantial agreement.

Cognitive variables

Table 1 Summary statistics comparing participants of the

1936 cohort who volunteered for the current MRI study
with non-participants
Variable

CIQ
EDU
OCC
RPM
DS
UCO
AVLT

Non-participants

t-test

Participants

Mean (SD)

Mean (SD)

P-value

37.8
2.97
5.05
34.0
41.6
13.3
58.4

434
278
276
263
249
238
256

45.1
3.25
5.43
37.4
45.2
13.9
59.0

227
225
224
223
208
211
220

50.001
0.156
0.054
50.001
0.001
0.284
0.629

(14.8)
(2.16)
(2.28)
(9.0)
(11.8)
(6.0)
(14.0)

(10.9)
(2.21)
(2.11)
(8.0)
(10.6)
(5.4)
(12.9)

CIQ = childhood intelligence; EDU = educational achievement score;

OCC = occupational coding; RPM = Ravens progressive matrices score; DS = digit
symbol test score; UCO = uses of common objects test score; AVLT = auditory
verbal learning test score; the latter four are cognitive ability tests taken in late life
(64 years of age).

Cognitive reserve variables

Most participants had educational attainment scores of 12, with a
mean score of 3.3 and only five participants had higher or professional degrees. Occupational grade ranged from 1 to 9 with a
mean of 5.4.

Statistical analysis and modelling

Table 2 shows bivariate (Pearsons) correlations demonstrating significant correlations (P 5 0.05) between childhood intelligence,
education, occupation and late life general intelligence factor g
and between head size and brain volume variables. There were
no significant relationships between cerebrovascular disease burden
(white matter hyperintensities) and childhood intelligence, education, occupation or brain volume measures. However, as expected,
there was a significant, negative correlation (r =  0.24; P 5 0.05)
between white matter hyperintensities and late life g. General
linear modelling with sex and childhood intelligence as fixed factors showed no additional significant relationships. In particular,
there was no relationship between education and any of imaging-derived variables. Preliminary analysis of Model 1 (Fig. 1) indicated that a better fit would be obtained if the error terms
associated with whole brain atrophy and total intracranial volume
were allowed to have a non-zero covariance. Model 2 (Fig. 2) is
similar to Model 1; however, this time brain atrophy is replaced by
hippocampal atrophy. Hippocampal atrophy is modelled as a latent
variable that directly affects left and right hippocampal atrophy.
Preliminary analysis of this model indicated that a better fit would
be obtained if the error terms associated with total intracranial
volume and hippocampal atrophy were allowed to have a nonzero covariance. The modification approach followed was that described by Joreskog and Sorbom (1996) and was done within AMOS
18. Following these modifications, both models demonstrated excellent fits to the data (chi-squared to degrees of freedom ratio 41.3,

Table 2 Bivariate correlations between putative influences on cognitive reserve and late life fluid intelligence
Variable

CIQ

EDU

OCC

L Hippo

R Hippo

BV

TICV

WMH

EDU
OCC
L Hippo
R Hippo
BV
TICV
WMH
g

0.46**
0.41**
0.03
0.03
0.01
0.01
0.13
0.49**

0.58**
0.10
0.07
0.11
0.08
0.12
0.44**

0.00
0.02
0.12
0.14*
0.06
0.33**

0.82**
0.64**
0.56**
 0.09
0.08

0.56**
0.46**
0.13
0.06

0.92**
0.01
0.03

0.06
 0.07

0.24**

*P 5 0.05; **P 5 0.01.

BV = brain volume; CIQ = childhood intelligence quotient; EDU = educational achievement; g = general intelligence factor; L Hippo = left hippocampal
volume; OCC = occupational coding; R Hippo = right hippocampal volume; TICV = total intracranial volume; WMH = total white matter hyperintensity
(Scheltens) score.

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

Among 249 Aberdeen 1936 Birth Cohort participants with MRI,

223 completed Ravens Progressive Matrices, 220 completed
Auditory Verbal Learning Test, 211 completed Uses of Common
Objects and 208 completed Digit Symbol. Neuropsychological
tests in late life are all highly and positively inter-correlated.
Principal components analysis was used to identify the first unrotated principal component g, which accounted for 47% of the
variance in the scores for all subjects. The component loadings

were Ravens Progressive Matrices = 0.67, Auditory Verbal

Learning Test = 0.72, Digit Symbol = 0.71 and Uses of Common
Objects = 0.63. The standardized variable was multiplied by 15
and added to 100 to create an Intelligence Quotient-type score,
which was entered into the analyses (Deary, 2003).

Cognitive reserve and neuroimaging disease burden

Brain 2011: 134; 36873696

| 3691

CIQ = childhood intelligence; EDU = educational achievement; g = general intelligence factor; OCC = occupational coding; TICV = total
intracranial volume; WMH = total white matter hyperintensity (Scheltens) score; eo, ee, et, eb and eg represent the error associated with
their respective variables. Thick correlation and b lines represent significant relationships (P 4 0.05).

Figure 2 Structural equation model with mean hippocampal volume as volumetric variable. CIQ = childhood intelligence;
EDU = educational achievement; g = general intelligence factor; Hippo Atrophy = latent variable of hippocampal atrophy; LH = reciprocal
of left hippocampal volume; OCC = occupational coding; RH = reciprocal of right hippocampal volume; TICV = total intracranial volume;
WMH = total white matter hyperintensity (Scheltens) score; eo, ee, et, eb and eg represent the error associated with their respective
variables; thick correlation and b lines represent significant relationships (P 4 0.05).

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

Figure 1 Structural equation model with total brain volume as volumetric variable. Brain atrophy = reciprocal of whole brain volume;

3692

| Brain 2011: 134; 36873696

Discussion
This study adds to our understanding of cognitive reserve by
showing that educational attainment, but not occupation, has a
measurable and positive effect on late life cognitive ability in
people aged 68 years, without dementia. Importantly, our model
includes the underlying influences of childhood intelligence and
the two core features of ageing brain. In addition, we demonstrate
that the positive contribution of education is greater than the
negative contributions of either cerebrovascular disease (measured
using Scheltens score) or Alzheimers disease (when estimated
from whole brain or hippocampal volumes) alone and similar to
their combined effect, illustrating how, in normal cognitive ageing,
reserve attained through education balances the effect of subclinical disease in late midlife.

Parameters of brain and cognitive

reserve
These results concur with those of the French three cities (3Cs)
study (Foubert-Samier et al., 2010), which found that education,
occupation and leisure activities contributed to cognitive (neural)
reserve and that education alone contributed to passive (brain)
reserve. Our results are also consistent with a recent report from
the Epidemiological Clinicopathological Studies in Europe (EClipSE)
collaboration, which found those with greater duration of education had both less dementia and heavier brains at the time of
death (EClipSE Collaborative Members, 2010). They also found
no relationship between duration of education and neuropathological measures of either Alzheimers disease or vascular
disease, supporting the hypothesis that the protective effect of

education acted through reserve rather than through common

factors underlying differences in both education and pathology.
The protective effect of educational attainment against dementia
has been demonstrated in the religious orders study, minimizing
the effect of other environmental influences (McDowell et al.,
2007). The inverse relationship between education and severity
of molecular and structural imaging surrogates of Alzheimers
pathology has been recognized for some time (Stern et al.,
1992; Bennett et al., 2003; Perneczky et al., 2006; Roe et al.,
2008). In a similar vein, Querbes et al. (2009) have argued, based
on follow-up data in mild cognitive impairment and Alzheimers
disease, that MRI-derived cortical thickness is a better (albeit more
expensive) method for diagnosis of these conditions than cognitive
testing because of the confounding effect of education. The contribution of engagement in leisure activities to cognitive function
in late midlife is established (Fabrigoule et al., 1995) and is related
to the diversity of activities, rather than the duration of any individual activity (Eskes et al., 2010).

White matter hyperintensities as a

biomarker of cerebrovascular disease
Cerebrovascular disease, seen as white matter hyperintensities on
brain MRI, is ubiquitous in older people and is known to be associated with reduced fluid intelligence, stroke, dementia and death
(Breteler et al., 1994; Leaper et al., 2001; Verdelho et al., 2007;
Debette and Markus, 2010). The negative association of white
matter hyperintensities with intelligence increases linearly with
age (Murray et al., 2011). White matter hyperintensities are
most prevalent and progress fastest in the frontal lobes (Sachdev
et al., 2007), are a recognized biomarker of vascular disease
(Manolio et al., 1994) and are associated with vascular risk factors
such as hypertension and impaired glucose tolerance (Murray
et al., 2005). They frequently co-exist with other neuropathologies
in older brains, most commonly those associated with Alzheimers
disease, and the presence of white matter hyperintensities in patients with Alzheimers disease should be regarded as the norm,
rather than the exception (van der Flier et al., 2007). While it is
possible that common genetic and/or environmental risk factors
may underlie both cerebrovascular and Alzheimers diseases, most
studies have found no association between white matter hyperintensities and hippocampal atrophy (Du et al., 2006) and white
matter hyperintensities are most likely to be coincidental to
Alzheimers neuropathology, with age as the major risk factor
for both pathologies (Greenberg et al., 2008). However, white
matter hyperintensities in posterior parietal and periventricular regions may be more specific to Alzheimers disease (Brickman et al.,
2009).

Hippocampal volume as a biomarker of

Alzheimers disease
Neuropathologically, the abnormal hyperphosphorylated intracellular neurofibrillary tangles of Alzheimers pathology progress in
characteristic stages from the entorhinal cortex to the neocortex
(Braak and Braak, 1991). Atrophy beginning in the entorhinal

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

root mean square error of approximation 40.03). Figure 1 shows

Structural Equation Model 1, which indicates that both childhood
intelligence and education have a large and positive influence on
late life g and that white matter hyperintensities have a negative
effect. Neither occupation nor whole brain atrophy has a significant effect on late life g in this model (see Table 3 for standardised
regression weights and significance). When whole brain atrophy is
replaced with hippocampal atrophy in Structural Equation Model 2
(Fig. 2), the positive influences are unchanged and both white
matter hyperintensities and hippocampal atrophy have a significant and negative influence on late life intelligence (see Table 4
for standardised regression weights and significance). These
models suggest that the negative effects on the left of the
models are in general balanced by the positive influences on the
right. It is possible that the patterns of results and significances are
brought about by modification of effects within the model as a
whole. Splitting the models into one positive and two negative
sub-models produced models that were good to excellent fits to
the data (available as online Supplementary material). The effects
and patterns of significances were unchanged from the parent
models. This indicates not only that the parent models are appropriate for the data, but that individual sections of the model are
also appropriate and that the pattern of balance observed is
authentic.

A. D. Murray et al.

Cognitive reserve and neuroimaging disease burden

Brain 2011: 134; 36873696

Table 3 Standardized regression weights, correlation coefficients and statistical analyses for structural equation
Model 1 (Fig. 1)

| 3693

Table 4 Standardized regression weights, correlation coefficients and statistical analyses for structural equation
Model 2 (Fig. 2)

Directional
parameters

Standardized regression
weights (b) (95% CI)

P-value

Directional
parameters

Standardized regression
weights (b) (95% CI)

P-value

CIQ!EDU
CIQ!OCC
EDU!OCC
Sex!TICV
Sex!Brain atrophy
OCC!g
EDU!g
CIQ!g
WMH!g
Sex!g
TICV!g
Brain atrophy!g

0.46
0.18
0.49
0.67
0.58
0.07
0.23
0.33
0.15
0.17
0.08
0.14

***
0.003
***
***
***
0.300
***
***
0.010
0.028
0.595
0.344

CIQ!EDU
CIQ!OCC
EDU!OCC
Sex!TICV
OCC!g
EDU!g
CIQ!g
WMH!g
Sex!g
TICV!g
Hippo atrophy!R
hippo atrophy
Hippo atrophy!L
hippo atrophy
Hippo atrophy!g

0.46
0.18
0.49
0.67
0.10
0.22
0.34
0.13
0.15
0.08
0.82

***
0.003
***
***
0.169
0.002
***
0.025
0.049
0.366

(0.36 to 0.55)
(0.05 to 0.30)
(0.38 to 0.61)
( 0.74 to 0.61)
(0.51 to 0.67)
( 0.06 to 0.21)
(0.10 to 0.38)
(0.23 to 0.46)
( 0.27 to 0.04)
(0.03 to 0.33)
( 0.36 to 0.18)
( 0.37 to 0.08)

Correlation coefficient (r)

P-value

CIQ!Sex
CIQ!WMH
Sex!WMH
eb!et

0.04
0.13
0.08
0.87

0.579
0.054
0.255
***

( 0.09
( 0.25
( 0.20
( 0.90

to
to
to
to

0.17)
0.00)
0.05)
 0.83)

CIQ = Childhood intelligence quotient; EDU = educational achievement;

g = general intelligence factor; OCC = occupational coding; TICV = Total intracranial volume; WMH = total white matter hyperintensity (Scheltens) score;
eb = error associated with brain atrophy; et = error associated with total intracranial volume, ***P 5 0.001.

cortex and hippocampi is recognized as a structural imaging feature of Alzheimers disease (Jack et al., 2002; Dubois et al., 2007)
and hippocampal atrophy is a risk factor for cognitive decline and
dementia in normal ageing (Ikram et al., 2010). Rate of hippocampal atrophy is thought to be more accurate than crosssectional measurement (Jack et al., 2004) and a meta-analysis of
rates of hippocampal atrophy in 595 people with Alzheimers disease and 212 controls found an annualized rate of hippocampal
atrophy of 4.66% in Alzheimers disease, compared with 1.41% in
controls (Barnes et al., 2009). Some studies have shown greater
preponderance for Alzheimers pathology to affect structure or
metabolism in the left hippocampus and/or left hemisphere in
Alzheimers disease (Gur et al., 1991), but this has not been a
consistent finding and depends on analytical strategy; for example,
correction for total intracranial volume and/or sex (Greenberg
et al., 2008). Clearly, hippocampal and temporal grey matter atrophy are found in cognitively normal older people, but their presence is predictive of dementia and death in follow-up of wellcharacterized normal cohorts (Ikram et al., 2010; Staff et al.,
2010). Here, we have used the reciprocal of hippocampal volume
as an indication of atrophy and as a parameter of Alzheimers
pathology. Using hippocampal volume as an indicator of passive
reserve would be a contrary, but equally valid design. Obviously, a
single measure of hippocampal volume does not indicate longitudinal atrophy and an alternative hypothesis would be that life-long
small hippocampal volume was a marker for low brain reserve and
an increased risk of dementia in late life.

0.98 (0.89 to 1.04)

0.20 ( 0.32 to  0.04)

***
0.008

Correlations

Correlation coefficient
(r) (95% CI)

P-value

Hippo atrophy!CIQ
CIQ!Sex
Hippo atrophy!Sex
Hippo atrophy!WMH
Sex!WMH
CIQ!WMH
Hippo atrophy!et

0.06 ( 0.07 to 0.19)

0.04 ( 0.09 to 0.17)
0.30 (0.15 to 0.39)
0.09 ( 0.00 to 0.24)
0.080 (  0.20 to 0.05)
0.13 ( 0.25 to 0.00)
0.49 ( 0.59 to 0.37)

0.342
0.579
***
0.131
0.256
0.054
***

CIQ = Childhood intelligence quotient; EDU = educational achievement; g = general intelligence factor; L Hippo = left hippocampal; OCC = occupational coding; R
Hippo = right hippocampal; TICV = Total intracranial volume; WMH = total white
matter hyperintensity (Scheltens) score; et = error associated with total
intracranial volume, ***P 5 0.001.

Strengths
We have studied a narrow age range sample that is very well
characterized over 410 years of follow-up. We have included a
valid measure of childhood intelligence and detailed measures of
late life intelligence allowing modelling of the effect of positive
and negative influences on life-long cognitive change. We have
included quantifiable measures of the two most common
brain pathologies in late life, namely cerebrovascular disease and
Alzheimers disease. We have also tested our hypotheses using
structural equation modelling, constructing logical and realistic structural models, which would not be possible using conventional regression-like approaches. The participants who volunteered for
recruitment to the Aberdeen 1936 Birth Cohort were more intelligent as children and as adults lived in more affluent neighbourhoods than those who did not volunteer, thus the relationships
demonstrated are unlikely to be spurious and are generalizable to
the wider population.

Weaknesses
Volume may not be the only measure of passive reserve and lack
of a significant relationship with volume does not preclude

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

Correlations

(0.36 to 0.55)
(0.05 to 0.30)
(0.38 to 0.61)
( 0.74 to  0.61)
( 0.03 to 0.23)
(0.08 to 0.37)
(0.23 to 0.46)
( 0.25 to  0.02)
(0.01 to 0.30)
( 0.25 to 0.10)
(0.68 to 0.86)

3694

| Brain 2011: 134; 36873696

is becoming more tangible with functional MRI studies beginning

to show evidence of a common reserve network in young people,
but not old, which could equate to cognitive reserve (Stern et al.,
2008) and with intellectual enrichment related to resting activity
in the default network (Sumowski et al., 2010). We have previously demonstrated, in a subgroup of the sample presented here,
that those showing poorer than expected late life intelligence relocate sites of activation during the inspection time task from
frontal to parietal lobes (Waiter et al., 2008). This is equivalent
to compensatory reserve, likely to be one of the earliest indicators of a failing brain and probably not a good thing.
The importance of understanding ageing and the complex interplay of multiple influences on cognitive ageing is clear
(Singh-Manoux and Kivimaki, 2010). Understanding how reserve
might be influenced to minimize the impact of neuropathologies
associated with dementia could have enormous public health implications (Brayne et al., 1998). This is a crucial prerequisite to
meaningful research in dementia and illustrates how life-long intellectual engagement can mitigate the negative impact of brain
pathology even on healthy ageing.
Future work should measure the contribution of more diverse
influences on cognitive reserve that might operate in early and
midlife, such as socio-economic conditions (Goldbourt, 2007)
and social relationships (Holt-Lunstad et al., 2010), which might
be modified through public education in order to have a positive
impact on the looming public health disaster that is dementia.

Acknowledgements
Dr Murray, Dr Staff, Dr McNeil, Dr Salarirad, Dr Ahearn and Ms
Mustafa are members of the Scottish Imaging Network (www
.sinapse.ac.uk). The authors would like to thank the participants
of the Aberdeen 1936 Birth Cohort, without whom this research
would not have been possible.

Funding
This work was supported by a grant from the Alzheimer Research
Trust (now Alzheimers Research UK). Data collection was
supported by funding from the University of Aberdeen
Development Trust.

Supplementary material
Supplementary material is available at Brain online.

References
Arbuckle JL. AMOS 7.0 Users Guide. Chicago: SPSS; 2006.
Barnes J, Bartlett JW, van de Pol LA, Loy CT, Scahill RI, Frost C, et al. A
meta-analysis of hippocampal atrophy rates in Alzheimers disease.
Neurobiol Aging 2009; 30: 171123.
Bennett DA, Wilson RS, Schneider JA, Evans DA, Mendes de Leon CF,
Arnold SE, et al. Education modifies the relation of AD pathology to

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

differences, for example, in receptor density. However, hippocampal atrophy is a well-accepted biomarker of Alzheimers pathology.
Cross-sectional volumes do not confirm atrophy and those with
smaller hippocampi may always have had smaller hippocampi,
possibly through less educational endowment with passive
(brain) reserve. Use of Scheltens scale for assessment of white
matter hyperintensities can be criticized as prone to observer variation and ceiling effects (van Straaten et al., 2006). However,
magnetic resonance images were scored by a single, experienced
observer and the total score employed in the analyses presented
here is likely to be an accurate measure of vascular burden.
Educational attainment is a relatively crude metric and does not
capture individual differences in intellectual activity and intellectual
engagement out with formal education. We have not captured
differences in reserve that may be endowed through social/leisure
activities, as found in the 3Cs study (Foubert-Samier et al., 2010)
and there is evidence that these contribute significantly and that
variety of activities is more important than time spent on any
individual activity (Eskes et al., 2010). It could be argued that
the way in which we have recorded educational attainment is a
weakness of our study since the data are not normally distributed
and its effect may not be linear. We compared those with high
and low educational attainment by splitting the data set into two
groups: 120 participants were defined at having low educational
attainment (no school leaving qualifications or a lower leaving
certificate); 105 were defined as high educational attainment
(higher school leaving certificate and above). Removing educational attainment from the models shown in Figs 1 and 2 and performing two (high and low) group structural equation models
(Arbuckle, 2006; Murray et al., 2011) with direct influences
(single headed arrows) fixed between the groups, with the exception of white matter hyperintensities!g, brain atrophy!g and
hippocampal atrophy!g, which were allowed to be different between the groups. These models produced good/excellent fits to
the data. We found no significant difference between the high
and low groups in terms of the imaging measures effect on g in
both modified models. This would suggest that the MRI detected
brain changes have the same effect on g independent of educational attainment. A further criticism of our models may be that
we do not allow for the possibility that education has a direct
causal influence on brain structural measures. This possibility was
not indicated by the initial correlations reported in the results. In
addition general linear models, adjusting for sex and childhood
intelligence, also found no association between education and
brain structural measures (left hippocampus, right hippocampus,
brain atrophy and white matter hyperintensities), which would
support its absence in the model structure. Similarly, occupation
may not be normally distributed and it is unclear if the relationships between occupation, education, childhood intelligence and g
can be approximated as linear, which is assumed in our models.
Our positive sub-models indicate good/excellent fits to the data.
However, this should be considered as an absence of evidence
for the inappropriateness of our model rather than evidence of
appropriateness.
We have used structural MRI variables as measures of subclinical, dementia-related neuropathology and not explored functional
MRI. Although remaining a theoretical construct, cognitive reserve

A. D. Murray et al.

Cognitive reserve and neuroimaging disease burden

| 3695

Holt-Lunstad J, Smith TB, Layton JB. Social relationships and mortality

risk: a meta-analytic review. PLoS Med 2010; 7: e1000316.
Hoyle RH, Panter AT. Writing about structural equation models. In:
Hoyle RH, editor. Structural equation modeling: concepts, issues,
and applications. Thousand Oaks, CA: Sage Publications; 1995.
p. 15876.
Ikram MA, Vrooman HA, Vernooij MW, den Heijer T, Hofman A,
Niessen WJ, et al. Brain tissue volumes in relation to cognitive function
and risk of dementia. Neurobiol Aging 2010; 31: 37886.
Jack CR Jr, Shiung MM, Gunter JL, OBrien PC, Weigand SD,
Knopman DS, et al. Comparison of different MRI brain atrophy rate
measures with clinical disease progression in AD. Neurology 2004; 62:
591600.
Jack CR Jr, Dickson DW, Parisi JE, Xu YC, Cha RH, OBrien PC, et al.
Antemortem MRI findings correlate with hippocampal neuropathology
in typical aging and dementia. Neurology 2002; 58: 7507.
Joreskog K, Sorbom D. LISREL 8: Users reference guide. SSI Scientific
software International. Hillsdale, NJ: Lawrence Erlbaum Associates;
1996.
Leaper SA, Murray AD, Lemmon HA, Staff RT, Deary IJ, Crawford JR,
et al. Neuropsychologic correlates of brain white matter lesions depicted on MR images: 1921 Aberdeen birth cohort. Radiology 2001;
221: 515.
Manolio TA, Kronmal RA, Burke GL, Poirier V, OLeary DH, Gardin JM,
et al. Magnetic resonance abnormalities and cardiovascular disease in
older adults. The cardiovascular health study. Stroke 1994; 25:
31827.
Matthews FE, Brayne C, Lowe J, McKeith I, Wharton SB, et al.
Epidemiological Pathology of Dementia: Attributable-Risks at Death
in the Medical Research Council Cognitive Function and Ageing
Study. PLoS Med 2009; 6: e1000180.
McDowell I, Xi G, Lindsay J, Tierney M. Mapping the connections between education and dementia. J Clin Exp Neuropsychol 2007; 29:
12741.
Mortimer JA, Borenstein AR, Gosche KM, Snowdon DA. Very early detection of alzheimer neuropathology and the role of brain reserve in
modifying its clinical expression. J Geriatr Psychiatry Neurol 2005; 18:
21823.
Murray AD, Staff RT, McNeil CJ, Salarirad S, Starr JM, Deary IJ, et al.
Brain lesions, hypertension and cognitive aging in the 1921 and 1936
Aberdeen birth cohorts. Age (Dordr). 2011 Advance Access published
on March 22, 2011, doi:10.1007/s11357-011-9233-5.
Murray AD, Staff RT, Shenkin SD, Deary IJ, Starr JM, Whalley LJ. Brain
white matter hyperintensities: Relative importance of vascular risk factors in nondemented elderly people. Radiology 2005; 237: 2517.
Perneczky R, Drzezga A, Diehl-Schmid J, Schmid G, Wohlschlager A,
Kars S, et al. Schooling mediates brain reserve in Alzheimers disease:
findings of fluoro-deoxy-glucose-positron emission tomography. J
Neurol Neurosurg Psychiatry 2006; 77: 10603.
Querbes O, Aubry F, Pariente J, Lotterie JA, Demonet JF, Duret V, et al.
Early diagnosis of Alzheimers disease using cortical thickness: Impact
of cognitive reserve. Brain 2009; 132: 203647.
Raven JC, Court JH, Raven J. Manual for Ravens progressive matrices
and vocabulary scales. London, UK: Lewis; 1977.
Rey A. Lexamen clinique en psychologie. Paris: Presses universitair de
France; 1964.
Roe CM, Mintun MA, DAngelo G, Xiong C, Grant EA, Morris JC.
Alzheimer disease and cognitive reserve: variation of education effect
with carbon 11-labeled Pittsburgh compound B uptake. Arch Neurol
2008; 65: 146771.
Sachdev P, Wen W, Chen X, Brodaty H. Progression of white matter
hyperintensities in elderly individuals over 3 years. Neurology 2007;
68: 21422.
Scheltens P, Barkhof F, Leys D, Pruvo JP, Nauta JJ, Vermersch P, et al. A
semiquantative rating scale for the assessment of signal hyperintensities on magnetic resonance imaging. J Neurol Sci 1993; 114: 7.
Singh-Manoux A, Kivimaki M. The importance of cognitive ageing for
understanding dementia. Age 2010; 32: 50912.

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016

level of cognitive function in older persons. Neurology 2003; 60:

190915.
Braak H, Braak E. Neuropathological stageing of alzheimer-related
changes. Acta Neuropathol 1991; 82: 23959.
Brayne C, Gill C, Huppert FA, Barkley C, Gehlhaar E, Girling DM, et al.
Vascular risks and incident dementia: Results from a cohort study of
the very old. Dement Geriatr Cogn Disord 1998; 9: 17580.
Breteler MM, van Swieten JC, Bots ML, Grobbee DE, Claus JJ, van den
Hout JH, et al. Cerebral white matter lesions, vascular risk factors, and
cognitive function in a population-based study: The Rotterdam study.
Neurology 1994; 44: 124652.
Brickman AM, Muraskin J, Zimmerman ME. Structural neuroimaging in
Altheimers disease: do white matter hyperintensities matter?
Dialogues Clin Neurosci 2009; 11: 18190.
Deary IJ, Whiteman MC, Starr JM, Whalley LJ, Fox HC. The impact
of childhood intelligence on later life: following up the Scottish mental
surveys of 1932 and 1947. J Pers Soc Psychol 2004; 86: 13047.
Deary IJ, Leaper SA, Murray AD, Staff RT, Whalley LJ. Cerebral white
matter abnormalities and lifetime cognitive change: a 67-year followup of the Scottish mental survey of 1932. Psychol Aging 2003; 18:
1408.
Debette S, Markus HS. The clinical importance of white matter hyperintensities on brain magnetic resonance imaging: systematic review
and meta-analysis. Br Med J 2010; 341: c3666.
Du AT, Schuff N, Chao LL, Kornak J, Jagust WJ, Kramer JH, et al. Age
effects on atrophy rates of entorhinal cortex and hippocampus.
Neurobiol Aging 2006; 27: 73340.
Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger-Gateau P,
Cummings J, et al. Research criteria for the diagnosis of alzheimers
disease: revising the NINCDS-ADRDA criteria. Lancet Neurol 2007; 6:
73446.
EClipSE Collaborative Members; Brayne C, Ince PG, Keage HA,
McKeith IG, Matthews FE, et al. Education, the brain and dementia:
neuroprotection or compensation? Brain 2010; 133: 22106.
Eskes GA, Longman S, Brown AD, McMorris CA, Langdon KD,
Hogan DB, et al. Contribution of physical fitness, cerebrovascular reserve and cognitive stimulation to cognitive function in postmenopausal women. Front Aging Neurosci 2010; 2: 137.
Fabrigoule C, Letenneur L, Dartigues JF, Zarrouk M, Commenges D,
Barberger-Gateau P. Social and leisure activities and risk of dementia: a prospective longitudinal study. J Am Geriatr Soc 1995; 43:
48590.
Foubert-Samier A, Catheline G, Amieva H, Dilharreguy B, Helmer C,
Allard M, et al. Education, occupation, leisure activities, and brain
reserve: a population-based study. Neurobiol Aging 2010. Advance
Access published on November 11, 2010, doi:10.1016/j.
neurobiolaging.2010.09.023.
Gold DP, Andres D, Etezadi J, Arbuckle T, Schwartzman A, Chaikelson J.
Structural equation model of intellectual change and continuity and
predictors of intelligence in older men. Psychol Aging 1995; 10:
294303.
Goldbourt U. Socioeconomic status in relationship to death of vascular
disease and late-life dementia. J Neurol Sci 2007; 257: 17781.
Great Britain Office of Population Censuses and Surveys. Standard occupational classification: coding index. London: HMSO; 1990.
Greenberg DL, Messer DF, Payne ME, Macfall JR, Provenzale JM,
Steffens DC, et al. Aging, gender, and the elderly adult brain: An examination of analytical strategies. Neurobiol Aging 2008; 29: 290302.
Guilford JP, Christensen PR, Merrifield PR, Wilson RC. Alternate uses:
manual of instructions and interpretation. Orange, CA: Sheridan
Psychological Services; 1978.
Gur RC, Mozley PD, Resnick SM, Gottlieb GL, Kohn M, Zimmerman R,
et al. Gender differences in age effect on brain atrophy measured by
magnetic resonance imaging. Proc Natl Acad Sci USA 1991; 88:
28459.
Hall CB, Derby C, LeValley A, Katz MJ, Verghese J, Lipton RB. Education
delays accelerated decline on a memory test in persons who develop
dementia. Neurology 2007; 69: 165764.

Brain 2011: 134; 36873696

3696

| Brain 2011: 134; 36873696

Staff RT, Murray AD, Ahearn T, Salarirad S, Mowat D, Starr JM, et al.
Brain volume and survival from age 78 to 85: the contribution of
Alzheimer-type magnetic resonance imaging findings. J Am Geriatr
Soc 2010; 58: 68895.
Staff RT, Murray AD, Deary IJ, Whalley LJ. What provides cerebral reserve? Brain 2004; 127: 11919.
Stern Y. What is cognitive reserve? Theory and research application of
the reserve concept. J Int Neuropsychol Soc 2002; 8: 44860.
Stern Y. Cognitive reserve. Neuropsychologia 2009; 47: 201528.
Stern Y, Alexander GE, Prohovnik I, Mayeux R. Inverse relationship between education and parietotemporal perfusion deficit in Alzheimers
disease. Ann Neurol 1992; 32: 3715.
Stern Y, Zarahn E, Habeck C, Holtzer R, Rakitin BC, Kumar A, et al. A
common neural network for cognitive reserve in verbal and object working memory in young but not old. Cereb Cortex 2008; 18: 95967.
Sumowski JF, Wylie GR, Chiaravalloti N, DeLuca J. Intellectual enrichment lessens the effect of brain atrophy on learning and memory in
multiple sclerosis. Neurology 2010; 74: 19425.

A. D. Murray et al.
van der Flier WM, Barkhof F, Scheltens P. Shifting paradigms in dementia: Toward stratification of diagnosis and treatment using MRI. Ann N
Y Acad Sci 2007; 1097: 21524.
van Straaten EC, Fazekas F, Rostrup E, Scheltens P, Schmidt R, Pantoni L,
et al. Impact of white matter hyperintensities scoring method on correlations with clinical data: the LADIS study. Stroke 2006; 37: 83640.
Verdelho A, Madureira S, Ferro JM, Basile AM, Chabriat H, Erkinjuntti T,
et al. Differential impact of cerebral white matter changes, diabetes, hypertension and stroke on cognitive performance among
non-disabled elderly. The LADIS study. J Neurol Neurosurg
Psychiatry 2007; 78: 132530.
Waiter GD, Fox HC, Murray AD, Starr JM, Staff RT, Bourne VJ, et al. Is
retaining the youthful functional anatomy underlying speed of information processing a signature of successful cognitive ageing? An
event-related fMRI study of inspection time performance.
Neuroimage 2008; 41: 58195.
Wechsler D. Manual for the Wechsler adult intelligence scale-iii. New
York: The Psychological Corp.; 1997.

Downloaded from http://brain.oxfordjournals.org/ by guest on February 11, 2016